VLA-Powered
On-Device VLA Autonomy
SynapTree™, VLA-Powered Autonomous Intelligence
Powered by an on-device VLA (Vision-Language-Action) architecture,
SynapTree™ perceives real-world environments, understands them through language,
and generates autonomous behavior — all without reliance on cloud connectivity.
Moving beyond traditional SLAM and LiDAR-based path planning,
SynapTree marks the beginning of a new generation of AI-native autonomous navigation,
engineered for real-world complexity and seamless robot deployment.
SynapTree delivers fully autonomous navigation powered by an on-device VLA architecture, enabling exploration, decision-making, planning, and execution locally — optimized for lightweight, low-latency deployment in real-world service environments.
Context-Aware Autonomous Navigation with VLA
Autonomous Driving Algorithm Based on VLM and Sensor Fusion
SynapTree™ integrates complex sensor data, including LiDAR, cameras, IMUs, and GPS, to precisely perceive the environment. It then uses its own VLA model to understand the driving context and generate an optimal driving path in real time.
On Device VLA [Vision-Language-Action]
SLAM-Free Navigation
Camera-Perception
Context-Aware Mobility
LLM-Integrated Pathfinding
SynaAI On Device VLA Architecture
SynaAI VLA, a lightweight solution based on SLM (SynaAI 3B, Quantized) and VLM, generates context and actions through vision inference, enabling safe and effective autonomous driving in dynamically changing environments.
VLMN(Vision-Language Model Navigation) & RFM
SynaAI® Stack, a patented solution implemented with proprietary algorithms for on-device RFM (Real-time Foundational Model) implementation, is an ultra-lightweight multimodal AI stack that handles real-time inference, interpretation, decision-making, and action execution based on Vision-Language-Action (VLA) entirely on local devices within an on-device environment. The VLMM (Vision-Language Multimodal) Framework built upon it integrates visual and auditory information through a multimodal encoder, converts it into prompts deliverable to the SLM via a Tokenizer, and transforms them into executable action contexts. The ROS2-based Robot Controller effectively integrates with existing Waypoint generation, Path Planning, and Behavior Control functionalities, leading to vision-based autonomous driving and robot action execution modules.
VLA (Vision-Language-Action) based Autonomous Driving Core
VLN (iSLAM) – Real-time AI Vision Object Detection, Population (density/crowd) extraction from multiple cameras generate VLA Datasets that reflect spatial information, driving constraints, risks, and weights, providing them through a real-time pipeline.
- iSLAM Planning: Path planning integrated with localization and mapping (including semantic costmap). Autonomous Driving NODE M. / SW TF – Through
- Nav2/SLAM/Controller configuration and middleware, the ‘semantic information’ generated by VLA is fed into iSLAM-based navigation’s costmap and policies, providing context-aware driving that considers humans, crowds, high temperatures, and unusual situations.
Synaptree® V3, Adaptive Exploration
Adaptive Exploration-Frontier Explore, which detects unknown areas in real-time and updates maps through an autonomous exploration algorithm (Frontier Explore),
provides dynamic map updating in unknown spaces without prior map data. It grounds inferred vision recognition data into spatial coordinates and offers an autonomous driving platform that controls and operates driving through language. Synaptree V3 performs self-exploration and mapping (Frontier) and recovery driving (Wall-follower) even in unknown spaces. It integrates VLA-based object/place information into map/odom/base_link coordinate systems and provides ‘language-spatial fusion’ autonomous driving, where an LLM converts natural language commands into path and action policies.”
The SynapTree-Explore Manager (BT/SMACC2) serves as the central orchestration unit, managing the overall autonomous exploration process. It interfaces with iSLAM / Mapping (RTAB-Map/iSLAM), which is responsible for real-time localization and simultaneous mapping. From the iSLAM output, the system proceeds to Frontier Explore (explore_lite/m-explore), an adaptive exploration algorithm that dynamically updates maps in unknown territories. Concurrently, Wall-follower (PID Local Controller) handles recovery driving scenarios. The SynapTree® V3, integrating these components, uses VLA-based vision recognition data, grounds it into spatial coordinates, and controls operation via language commands. The LLM (Language-Spatial Fusion) converts natural language instructions into actionable path and behavior policies, enabling context-aware driving in diverse, even unknown, environments
Verifiable and Scalable Simulation
& Open Development Environment
Verifiable and Scalable Simulation and Open Development Environment for Autonomous Driving Algorithms – Synaptree™ V4, with its sophisticated integration of active exploration, autonomous driving auto-switching, separated updates for dynamic/static changes, VLA (Vision-Language-Action) semantic-based avoidance, destination control, and constraints, provides Adaptive Exploration & Dynamic Mapping for robot operation in environments with maximized dynamic changes. It offers various simulation methods during the development phase, including a ROS2 port of explore_lite for m-explore-ros2 and Nav2 + wavefront-frontier, as well as frontier exploration combined with the Nav2 waypoint follower.
SynaAI Edge VLA for Next-Generation Spatial Reasoning and Path Planning
SynaAI V2 performs real-time spatial perception and dynamic path planning through an on-device Vision-Language Model (VLM) designed to combine natural language prompts and visual data to understand, analyze, and respond to complex environments in real-time, enabling autonomous driving even in changing situations.
The Synaptree solution, an autonomous driving platform based on SynaAI V2, will commence its demonstration service at Incheon International Airport in February 2025. In collaboration with Samsung Electronics, it has become Korea’s first commercialized VLA autonomous driving AI platform and is being adopted in various next-generation robot projects. Integrit continues to expand and advance its next-generation on-device AI platform, which combines scalability and high performance.
Next-Generation Autonomous Driving Advanced with On-Device AI
Integrit’s next-generation on-device platform is at the forefront of advanced AI robotics and AI transformation
Integrit’s on-device AI-based autonomous driving, robotics, and spatial intelligence solutions are being deployed and operated in real-world environments as core technologies. The VLA (Vision-Language-Action) based autonomous driving solution, Synaptree® V3, has been validated across various AI robot platforms, including advanced AI service robots, smart mobility, Boston Dynamics Spot, and home agent robots. It is being introduced as a next-generation AI technology in real-world settings such as Incheon International Airport, Samsung Electronics, The Hyundai, Parc.1, and HL Mando.
0
누적주행거리(Km)
0
누적주행시간(Hours)